The Increasing Importance of Geographical Proximity in Technological Innovation:

Transcription

1 The Increasing Importance of Geographical Proximity in Technological Innovation: An Analysis of U.S. Patent Citations, Jung Won SONN and Michael STORPER Sonn: PhD Candidate, Dept. of Urban Planning, UCLA Storper: Professor of Economic Sociology, Institute of Political Studies (Sciences Po), Paris and Professor, Dept. of Geography, London School of Economics Corresponding author: Jung Won Sonn School of Public Policy and Social Research 3250 Public Policy Building UCLA Los Angeles, California USA Paper prepared for the Conference: What Do we Know about Innovation? in Honour of Keith Pavitt, Sussex, November 2003 Abstract In spite of increasing global flows of ideas, capital, goods and labor, recent research in urban economics and economic geography suggests that geographical proximity between innovators may be important to technological innovation. Many authors also claim that the rise of a knowledge-based economy and changes in the organization of the innovation process have actually increased the value of such proximity to innovation. Yet there is little empirical research on whether this latter proposition is valid. Using patent citation data, this paper investigates changes over time in inventors dependence on locally-created knowledge. We find an increasing tendency for inventors to cite local patents at three geographical levels: the national, the state and metropolitan levels. This implies that inventors increasingly use domestic knowledge more than foreign knowledge, in-state knowledge more than out-of-state knowledge, and knowledge from the same metropolitan area more than knowledge from outside. Thus, proximity in the creation of economically-useful knowledge appears to be becoming even more important than was previously the case.

2 Increasing Importance of Proximity in Innovation 2 Keywords: proximity, patent citation, localized knowledge spillover, innovation, invention The Advantage of Geographical Proximity to Innovation: Not Dead Yet or Increasingly Important? The last few decades have witnessed great increases in global flows of ideas, capital, goods and labor (Dicken 1992; Scott 1998; Held et al. 1999). In particular, information, at least in its codified forms, flows without much friction from one place to another. Almost weightless, it can reach most parts of the world with the speed of the internet or a courier service, and at low cost. Furthermore, the rapid transition to a knowledge based economy has increased the incentives to exploit knowledge produced in other parts of the world, since knowledge rents are increasingly important in contemporary competition. However, information is not equivalent to economically-useful knowledge, and the latter may experience considerable friction to distance. Along these lines, urban economists and economic geographers have suggested that geographical proximity between the people and organizations who produce knowledge may still be an important advantage in the production of economically-useful innovations ( Acs, 2002; Dicken, 1992; Feldman, 1994; Storper, 1997). Innovations with a high technological content have been shown to be facilitated by the physical co-presence of key scientists and frequent interaction between related workers (Saxenian, 1994; Zucker, Darby and Armstrong,

3 Increasing Importance of Proximity in Innovation ). This would be the case if certain kinds of knowledge flow were subject to constraints related to the cost or feasibility of covering distance, or both ( Acs, 2002; Audretsch and Feldman, 1996; Feldman and Florida 1994; Foray 1991; Glaeser 1999; Marshall 1890; Patel and Pavitt, 1991; Porter 2000; Smith 1999; Storper 1997). Indeed, a broad literature has emerged which sees innovation as a systemic process which depends on multiple interactions, which are structured around nodes and channels of various types, each involving different types of relations between innovating agents, and with different degrees of intensity and hence costs and feasibility of carrying them out over geographical space (Breschi, Lissoni and Malerba, 2003). Along these lines, Storper and Venables (forthcoming, 2004) analyze local direct face-to-face contact, showing that it is not only an efficient technology of communication, but also a means for reducing moral hazards in relationships which are extremely uncertain, such as in the formation of joint projects to develop new knowledge, where complete contracts are impossible and bureaucracies too costly. Storper (1997) argues that relationships, which depend in varying degrees on geographical proximity, are the key untraded interdependencies that allow actors to coordinate in order to innovate. Even more intriguing is the suggestion that the importance of proximity be actually increasing with the advent of an increasingly knowledge-based economy. In this view of things, contemporary information and communication technologies have permitted a much more flexible and rapid interaction between product and process change in many industries, necessitating constant innovation and introducing new islands of complexity in the economic process which require dense, person-to-person

4 Increasing Importance of Proximity in Innovation 4 communication among potential innovators. Added to this is the opening up of new frontiers in basic technologies such as microelectronics, the internet and biotechnology, where constant innovation is at the heart of those industries performance. Still others claim that both long-distance and local interactions should intensify in the knowledgebased economy, with multinational firms increasingly embedding themselves in localized innovation systems and acting as long-distance links between such nodes in global networks (Verspagen and Schoenmakers, 2002). Yet this apparent continued importance of proximity seems paradoxical in today s world and is highly contested by certain analysts. If it is true that we are getting closer and closer to a condition where the friction of distance is dying, then the importance of proximity should be decreasing (Cairncross, 1997). The problem is that any claim that the importance of proximity is rising cannot be arbitrated on the basis of existing research, because most studies of knowledge flows and related topics have employed panel data. Such data sets cannot track historical changes in the geography of knowledge flows; they can only measure it at one point in time. This paper attempts to fill some of this gap using an historical patent citation data base which is geographicallyspecific. With the aid of the Jaffe, Trajtenberg, and Henderson Control (JTH Control), we investigate the historical changes in inventors dependence on locally created knowledge at three geographical levels: country, the U.S. state, and the U.S. metropolitan areas. The paper comprises four sections. It begins by reviewing empirical studies on the importance of proximity and formulates two main hypotheses. The next section explains data and methods that the empirical study will employ. Then it reports the

5 Increasing Importance of Proximity in Innovation 5 results of the empirical study, and concludes by summarizing the findings and a considering possible implications. The Dearth of Empirical Research on the Changing Role of Geographical Proximity Over Time Since the late 1980s, a number of studies have found evidence of the benefit of proximity in technological innovation. Employing aggregate data mostly at the state level and occasionally at the metropolitan level, this body of literature estimates a knowledge production function and finds some evidence for the importance of proximity. The pioneering work is that of Jaffe (1989). He finds that for each U.S state, the number of patents by businesses is positively correlated with the R&D expenditure of universities located in the same state. Based on this finding, he infers, with caution, that there may be knowledge spillovers from universities to businesses. Similar results were obtained by Acs, Audretsch and Feldman (1992) who replicated Jaffe (1989) with different data: the Innovation Count from the Small Business Innovation Data Base. Kelly and Hageman (1999) control for the distribution of production so as to isolate the effects of knowledge spillovers. They hypothesize that in the absence of knowledge spillovers, the distribution of R&D activities (measured by the number of patents) should be analogous to the distribution of production facilities (measured by the number of workers). Employing a quality ladder model with patent counts at state level, they find that patenting is geographically clustered even after controlling for the location

6 Increasing Importance of Proximity in Innovation 6 of employment. These three studies employ the state as the geographical scale of research. However, as Breschi and Lissoni (2001) rightly point out, the state is not an appropriate geographical unit for measuring knowledge spillovers because US states are administrative boundaries which do not neatly coincide with the boundaries of economic interactions. Moreover, economic policy at the state level is much less significant than that of the federal government. To complicate matters, US metropolitan areas sometimes extend across state boundaries. More appropriate scales are the country and the metropolitan area. A country coincides with boundaries of culture, language, public policy and currency, and thus constitutes a geographical scale of considerable importance. This notion is corroborated in numerous studies of trade, which show that national boundaries matter, even for the most ordinary kinds of trade and even when there are few language and customs barriers (McCallum, 1995). By contrast, even though a metropolitan area does not coincide with boundaries of culture, public policy and currency, it is nonetheless a functioning geographical scale of intensive daily economic interaction. For scientists and engineers, a metropolitan area roughly coincides with the geographical limit for meeting other scientists and engineers regularly and at low cost via face-to-face contact. Therefore, employing the state as the geographical unit is a serious limitation of abovementioned papers. In addition to the problems related to geographical scale, previous studies neglect the effect of spatial interaction. As Lim (2003) and Smith (1997) point out, the spatial reach of knowledge spillovers is not always confined to one geographic unit,

7 Increasing Importance of Proximity in Innovation 7 whether it be country, state or metropolitan area. Nearby units should be at least partial beneficiaries of knowledge spillovers. Anselin, Varga and Acs (1997) respond to both of these problems. Building upon Jaffe (1989), they make two contributions. On one hand, for the first time, MSA level data are used; on the other hand, a spatial lag variable was added to the Jaffe (1989) knowledge production function so as to capture the effect of interaction between spatial units. They then find evidence of knowledge spillovers at both state and MSA levels. Another body of literature attempts to prove the existence of localized knowledge spillovers by showing that the technological characteristics of a place influence the technological characteristics of a firm located in that place. Patel and Pavitt (1991) reported that production of technology remains bounded by space in the sense that multinational firms are strong in patenting in the areas in which their home national economies are generally strong in innovation, and that both are highly correlated to the industries in which a particular economy is strongly export-oriented. This led them to conclude that there was a strong association between the organizational capacities of major multinationals and the national systems of innovation of their home countries. Baptista and Swann (1998) apply similar reasoning to the sub-national level. They test whether firms in well-developed industrial clusters innovate more actively. They use a UK innovation database composed of 248 firms between , finding positive correlation between a firm's innovativeness and the region s employment in the same industrial sector. Using 1981 CURDS survey, Baptista (2000) shows that the presence of the early adopters of new technology is positively correlated with the overall efficiency of knowledge flow in that region. He estimates a duration model to find

8 Increasing Importance of Proximity in Innovation 8 determinants of adoption of microprocessors and computer numerical control (CNC) devices, and concludes that the number of previous adopters of the technology in the region is positive and significant. These studies find evidence of local knowledge spillover through correlations between the innovativeness of a region or a firm and that of its surrounding area. While acknowledging their value, it should also be pointed out that this evidence is at best indirect. Due to the aggregate nature of their data, they cannot study individual points of knowledge spillover. An alternative is to use data on the individual interactions that might indicate the presence of knowledge spillover. Jaffe, Trajtenberg and Henderson (1993) pioneer this method, using patent citation data to track flows of information. After controlling for the existing distribution of inventive activities, they find that inventors tend to cite other patents from within the same geographical unit, be it a country, a state or a Metropolitan Statistical Area. Yet though this represents a considerable advance, it still uses panel data and therefore cannot illuminate temporal change in the importance of proximity. In this light, none of the research cited thus far can indicate whether the importance of proximity has been increasing or decreasing toward possible extinction, as so many death of distance authors suggest. The rare studies that investigate temporal changes in knowledge spillover include Jaffe and Trajtenberg (1996) and Johnson Siripong and Brown (2002). Jaffe and Trajtenberg (1996) record the application year of cited patents and investigate changes in the locations of citing patents over time. They find that in the early years after a patent is approved, citations are made disproportionately by inventors in the same country (a

9 Increasing Importance of Proximity in Innovation 9 localization effect). However, in later years, the proportion of foreign citations increases. This suggests that knowledge becomes available to people across long distance only after some time, thereby confirming the benefits of proximity in the production and timely acquisition of cutting-edge knowledge. However, they do not investigate whether the localization effect of today is stronger than the past, as we do in the present research. Johnson, Siripong and Brown (2002) attempt to find whether the distance between coauthors of patents and between a citing patent and a cited patent have increased over time. Controlling for the effects of technological fields and assignee types, they find so. Johnson, Siripong and Brown (2002) investigate physical distance itself rather than the economic or innovation effects of proximity. They define distance as simple physical distance between two points. Thus, in their model, 2,000 miles is twice as distant as 1,000 miles. With today s transportation and communication, however, 1,000 miles and 2,000 miles are not appreciably different in the sense that they are within in a day s flying distance, with a total travel time difference which is not enormous. Moreover, the economic and innovation benefits of proximity are not linear to distance but instead resemble the shape of a Gaussian curve. To be within walking distance is very different from being slightly further away. Likewise, partners within driving distance can exchange their information face-to-face at least once a day but people even slightly further away cannot and their marginal costs of contact rise very steeply in comparison to those where an overnight stay is not required. Also important are social, political and psychological boundaries. In this respect, as we shall see later on, metropolitan areas and national boundaries delimit discontinuous scales (i.e. going

10 Increasing Importance of Proximity in Innovation 10 beyond whose boundaries involve important non-linear social, political and psychological costs). Therefore, distance should be defined as a discontinuous variable with distinctive threshold values. Furthermore, Johnson, Siripong and Brown (2002) do not control for the existing distribution of inventive activities. Therefore, some of the decreasing importance of distance they find is due to the long-term historical diffusion of inventive activities from the Northeastern region to the South and the West (which however they do report in another paper (Johnson and Brown 2002). Two Hypotheses The preceding discussion enables us to formulate the two hypotheses which motivate the research reported here. First Hypothesis: the importance of local knowledge flows, as measured by patent citations, is increasing over time. Our first hypothesis is that proximity is increasingly important for the exact same reason that journalists argue that distance no longer matters. When the global market is integrated and flows of goods are faster than ever, businesses face competition from elsewhere and are forced to innovate continuously. In such a situation, proximity to sources of knowledge could be an important asset. Because the products of a particular place can reach the world market, there is competition in many local markets between the same or similar products from different places. To cope with stronger competition, firms must innovate faster than their competitors. This strategy, in turn, shortens the life cycles of technologies. Due to these

11 Increasing Importance of Proximity in Innovation 11 shortened life-cycles of technology, the flow of information accelerates, often making it impossible to make the investments in codification and standardization that would in turn make it possible to transfer knowledge efficiently over long distances. One possible way of coping with this situation is to keep all the necessary knowledge sources in-house. This turns out not to be a good solution because organizational rigidity and the uncertain trajectory of technology oftentimes makes in-house innovation less efficient (Lundvall, 1992; Powell, 1990). Another way is to build stable bridges with other nearby firms so as not to miss any valuable information (Cooke and Morgan 1993; Scott 1996). The largest firms attempt to combine the two, inserting themselves into localized innovation environments, but also acting as network structures between such regional nodes at a global scale (Dunning, 2000; Verspagen and Schoenmakers, 2002). Therefore, improvement in communication and transportation technologies and the integration of the world market paradoxically result in a stronger localized flow of knowledge (Leamer and Storper, 2001). Second hypothesis: dependence on local interactions among innovators differs by assignee types. Different types of patent assignees should have different degrees of dependence on local knowledge spillover. For example, larger firms are probably better at overcoming frictions of distance. With branches in many different parts of the world, they can substitute intra-organizational proximity for geographical proximity. In addition, they have the resources to send researchers and engineers around the world to acquire new information created via routinized relationships in distant locations. They usually have the means for greater investments, both technological and organizational, in communication, hence reducing uncertainty while raising

12 Increasing Importance of Proximity in Innovation 12 understanding and lowering the marginal costs of efficient communication at a given level of complexity. Some authors have claimed, for example, that large firms can invest in the creation of long-distance epistemic communities that allow them to overcome the need for proximity (Hakanson, 2003; Lissoni 2003). In contrast, individual inventors, having less resource to access distant knowledge, can be expected to depend on local knowledge spillover to a greater extent than big firms or organizations. Method Following the major recent studies on knowledge spillover such as Jaffe, Trajtenberg and Henderson. (1993), Jaffe and Trajtenberg (1996), Jaffe and Trajtenberg (2002), Johnson, Siripong and Brown (2002), Koo (2002), Maurseth and Verspagen (2002) and Stolpe (2002), we will use patent citations as proxy of knowledge spillover 1. A U.S. patent record documents previous patents that are relevant. These recognitions or citations are regarded as a noisy indicator of interactions that lead to invention and innovation (Breschi, Lissoni and Malerba, 2003; Hall, Jaffe and Trajtenberg, 2001). By comparing the address of citing patent and cited patent, one can see whether citations are 1 Two major limitations of patent citation data are recognized by scholars interested in innovation. Firstly, as Feldman (1994) points out, some patents are not commercialized and thus do not have any impact on market economy. However, when the topic of research is flow of knowledge rather than direct economic impact, it would ultimately be important to include uncommercialized patents because they might be sources of inspirations to future patents that have great impacts on economy. Secondly, there are limitations of citation as a proxy of knowledge flow because patent examiners in the U.S. Patent and Trademark Office can add citations to past patents which they think relevant (Breschi, Lissoni and Malebra 2003 p.4). If added by the examiners, a citation does not represent actual flow of knowledge. However, as far as this error does not have systematic correlation with independent variables, it does not affect the research findings (Stolpe 2002).

13 Increasing Importance of Proximity in Innovation 13 locally made or not. However, one cannot simply take distribution of citations as distribution of knowledge spillover. Since R&D activities are unevenly distributed throughout the United States, cited patents should be unevenly distributed as well. Therefore, unless we control for the geographical unevenness of R&D activities, disproportionate citation of local patents might merely reflect the uneven geographical distribution of R&D activities. In this sense, only when citations are more highly localized than is patenting activity as a whole can it be said that there is localization in knowledge spillovers. To this end, this paper employs Jaffe, Trajtenberg, and Henderson s Control (JTH Control). To construct a control variable, Jaffe, Trajtenberg and Henderson (1993) connected a patent with a control patent that belongs to the same patent class and has the same application year and closest grant date. They then counted the number of cases where a patent s geographic location coincides with that of the control patent. The frequency of this coincidence is a proxy of the existing distribution of R&D activities. We count the number of incidents where citing and cited patents were from the same geographic unit, or what we call Total Local Citation. This procedure will be repeated for each year. In so doing, we will exclude citations in which the citing patent and the cited patent were assigned to the same assignee (self-citation). That is because self-citation is probably more the result of intra-organizational rather than geographically proximate flows of information. Net Local Citation Percentage (NLCP) was then calculated according to the following formula:

14 Increasing Importance of Proximity in Innovation 14 Net Local Citation Percentage = Total Local Citation Percentage JTH Control If NLCP at a certain geographic level is greater than zero, it suggests that inventors disproportionately cite past patents from the same geographical unit after controlling for the existing distribution of inventive activities. In other words, in the process of creating new technology, scientists and engineers are more dependent on knowledge created by other people in the same geographical unit than on knowledge in other units assuming that existing inventive activities are evenly distributed across geographical space. If NLCP at a given geographic scale is less than zero, it means that inventions are disproportionately citing past patents from geographical units other than their own after controlling for the existing distribution of inventive activities. This implies that, when creating new technology, scientists and engineers are more dependent on knowledge created by people in other geographical units, again assuming that existing inventive activities are evenly distributed. By comparing NLCPs of multiple years, we can determine whether the localization in knowledge spillover is becoming stronger or weaker. Data and Exploratory Analysis The main data set for this research comes from The NBER Patent Citations Data Files. This database contains information about patents, their inventors and citation records (Hall, Jaffe and Trajtenberg 2001). For the present research, we used data from application years 1975 to 1997.

15 Increasing Importance of Proximity in Innovation 15 We connected each patent with all the patents cited in its application. One pair consisting of a citing patent and a cited patent is defined as one unit for analysis. These pairs were then grouped according to the application year of the citing patent, and in this way cohorts for each year were generated. If inventors in a certain year are dependent upon geographically proximate sources of knowledge, patents in that year should disproportionately cite patents from the same geographical area. While Jaffe, Trajtenberg and Henderson (1993) grouped patents according to the application year of cited patents, we grouped them according to the citing patent because we are interested in determining whether the today s inventors are citing nearby patents than did inventors in the past. This process of data construction involves truncation problems at both ends. The address of the inventor at the metropolitan level is known only for patents granted between January 1, 1975 and December 31, Since the review process may take many years for some patent applications, approval decisions have not yet been made for a considerable portion of patent applications in the years just prior to Therefore, we removed patents applied for in 1998 and Even so, higher portion of patent applications made in later years is still under review than that of earlier years. At the opposite end of the data set, truncation is caused by citation lag. Jaffe, Trajtenberg and Henderson (1993) noted that the time lag between a citing patent and a cited patent varies significantly. Citations of a certain patent made within one or two years after the application is made are few. The third year onward, relatively stable numbers of citations are made each year until the eleventh year (See Figure 1, p. 587, Jaffe, Trajtenberg and Henderson, 1993). However, we are not able to estimate descriptive statistics of citation lag because patents will continue to receive citations in

16 Increasing Importance of Proximity in Innovation 16 the future. Furthermore, the dataset contains address information of patents granted only since This means that addresses of patents granted prior to 1975 cannot be found even if the citations to those patents are made between 1975 and Due to this problem, for earlier citing cohorts substantial portions of citations are made to patents granted prior to For example, for 1975 citing cohorts, out of 335,499 citations made, 109,943 or 32.8% were citations to patents granted prior to These figures are 31.7% for 1976, and 30.0% for 1997, but fall to 8.0% in 1997 and 7.8% in To compensate at least partially for this problem, we removed citations from the database for which the lag is more than seven years. We chose seven years because it is safely after the peak in the distribution of citations. More than half of the citations should be made within seven years after application. In this way, data sets of 1982 to 1997 have citations for which the lags are zero to seven years. However, this measure is only partially effective for citations between 1975 and The longest citations lag in 1981 is six years rather than seven because 1974 patents are not in the database. The longest lag for 1980 citations will be five years, for 1979 citations four years and so forth. Therefore, in the first years covered by the database, an increasingly smaller proportion of the citations actually made are in the database. We cannot know with any certainty whether the truncation problem generates bias, though if the analysis of these cohorts generates results which are significantly different from those of other years, it could be due to truncation. In any even, this procedure leaves 23 cohorts of citations from 1975 to The 23 cohorts were further categorized by assignee types. Using the assignee information of the citing patents, we organized all citations into five categories;

17 Increasing Importance of Proximity in Innovation 17 Individuals, the Federal Government, Universities, the Top 500 most innovative firms and Other firms. There are also patents which do not have assignees. We labeled these patents unassigned and treated as belonging to one separate type of assignees. As used here, assignee is a natural person or a legal person to which the legal ownership of the intellectual property rights have been transferred from the inventor. When a patent has no assignee, the inventor retains the intellectual property rights. If the inventor worked for a firm, a university or the federal government, the intellectual property rights are automatically transferred to the institution by which s/he is employed. As a result, the patents lacking assignees are very likely to be those developed by independent inventors. Some of the independent inventors may be hiring limited numbers of research staff while others conduct their research in their garage-lab alone or with a few friends. In this sense, it is safe to assume that the patents unassigned to anybody are those developed by the smallest entities. On the other hand, individual as an assignee type is the case when the intellectual property rights are transferred to an individual from the inventor. Since a patent developed by a firm is less likely to be sold to an individual, it is reasonable to assume that patents assigned to individual are the ones developed by as small entities as those unassigned. The technological or other differences between these two assignee types are not clear. Patents assigned to University, Firms or the Federal government are highly likely to be the ones developed by researchers working for those institutions. Among university assignees, the University of California is ranked top with 2768 patents granted between 1975 and The Massachusetts Institute of Technology and University of

18 Increasing Importance of Proximity in Innovation 18 Texas follow U.C. with 2151 and 1007 patents respectively. Among non-us universities, the University of British Columbia is ranked first with 229 patents and ranked 24 th overall. Hebrew University of Jerusalem is ranked 2 nd among foreign universities and 39 th overall. To our surprise, no Japanese, German or French universities are ranked among the top fifty (see Appendix 1). The Top 500 firms are those to which the most patents are assigned. This group of firms includes IBM, ranked first with 26,340 patents, and General Electric, ranked second with 25,868 patents. Hitachi is ranked first among non-u.s. firms and third overall. As expected, Canon, Toshiba, Eastman Kodak, AT&T, Du Pont, Motorola, Mitsubishi, Siemens and other well-known innovators are ranked highly (See Appendix 2). Other Firms include the firms that have less patents granted between 1975 and 1999 than the top 500 firms. Some of the Other Firms have as many patents as 467 but many have only one patent. Figure 1. Share of Each Assignee Category in Citations Figure 1 shows the share of each assignee category among all the citations made. The share of top 500 innovative businesses in overall citations decreased from 61% in 1975 to 24% in This result is consistent with the frequently advanced notion that the innovation process has been reorganized since the 1970s, away from the so-called linear innovation model in which large firms were expected to internalize many steps in the R&D/innovation process, toward an innovation process involving external, or interorganizational relationships between many different actors, and involving a back-and-

19 Increasing Importance of Proximity in Innovation 19 forth movement between science, R&D and commercialization. Such a change should involve just the kind of result we report, i.e. a decline in the percentage of patenting by the biggest firms. Other noticeable changes include a rise of university patenting (0.4% in 1975 to 1.4% in 1997) and decline of both the federal government (2.4% to 0.5%) and unassigned (15.7% to 12.0%). These changes may stem from the increasing commercialization of R&D. Universities are becoming more and more active in commercializing their inventions and thus attempt to protect their intellectual property rights by patenting their inventions. Independent inventors should attempt same thing. Since they are not able to transform their inventions into commercial products, they are more likely to sell their intellectual property right to an entrepreneurial entity who can. Furthermore the increasing investments required for successful R&D may leave less room for individual independent inventors. 2 Findings: The Importance of Proximity Increases over Time Figure 2 shows historical changes of NLCP at three geographical levels. At all three levels, the NLCP increases. At the country level, NLCP increases from 2.3% in 1975 to 11.4 % in At the state level, 1.9% to 5.3%, at the MSA level, 0.9% to 4.3% during the same period. 2 There is a debate over whether this represents a capital market failure, however, especially in those countries which have weak venture capital sectors.

20 Increasing Importance of Proximity in Innovation 20 Figure 2. Net Local Citation Percentage (Distribution of Citation Matches after Controlling for Existing Distribution of Inventive Activities) at Three Geographic Levels. The increase in citation activity at the national level is very clear. This result tends to confirm the position of Lundvall (1992), and Nelson (1993), who hold that national systems of innovation generate distinctive national trajectories of technological development, even in the face of considerable international technology spillovers and imitation (see also Cantwell and Iammarino, 2001). While globalization accelerates the exchange of products, a firm tends to have centers of R&D in its home country where it has better access to information sources and dense relational networks (Patel and Pavitt, 1991). Our research suggests that these effects are likely to remain in place and are perhaps intensifying over time. Proximity effects appear to be at work at the metropolitan level as well. This finding is consistent with the arguments made in Acs (2002), Braczyk, Cook and Heidenreich (1998), Cooke and Morgan (1998), and Fischer et al. (2001), that is, that systems of innovation exist not only at the national level but also at the metropolitan level. Most importantly, it shows that this effect has been increasing in recent years, to the point where metropolitan level interactions represent a non-negligible share of patent citations, at least in the USA. The state level findings are more puzzling. Our findings are consistent with prior research on the existence of knowledge spillovers at the state level, but we cannot say for sure whether the state is actually a significant geographical unit of knowledge spillover or the apparent significance is merely a lagged effect of knowledge spillovers

21 Increasing Importance of Proximity in Innovation 21 that take place within metropolitan areas within states. To determine whether there are knowledge spillovers at the state level, the effects of metropolitan knowledge spillovers must therefore be removed. To achieve this, we removed all cases where there is either a citation match or a control match at metropolitan level. This generates a hypothetical dataset where there are no metropolitan level knowledge spillovers and no metropolitan level of coincidental match of inventive activities. With this hypothetical database, we can then calculate the NLCP of each year at the state level. Figure 3. Net Local Citation Percentage at State Level after Removing Metropolitan Citation Matches As shown in Figure 3, after removing the lagged metropolitan effects, the localized knowledge spillover at the state level is very close to zero. The NLCPs of most years are over zero and they are statistically significant, but the overall size of NLCPs are very small (See Table. 1). Considering that some portion of knowledge spillover happens from a metropolitan area to nearby metropolitan areas, (which we cannot estimate at this point) there is insignificant level of citation activity at state level. This implies that if data are available at the metropolitan scale, then there is little to be gained by further analysis at state level. Figure 4. Net Local Citation Percentage by Assignee Types at Three Geographic Levels. Figure 4 shows changes for each assignee category. A prominent feature of Figure 3 is that the NLCP increases for all assignee types at all three levels except the

22 Increasing Importance of Proximity in Innovation 22 federal government at state and metropolitan levels. Although the NLCPs of unassigned, individual and university fluctuate due to the small number of cases, statistical comparison of the beginning years and ending years shows the increase to be statistically significant. We calculated weighted averages of the NLCP of the first five years and the last five years, so as to smooth random fluctuations; we then conducted t- tests with these two ratios. For sixteen cases of increase in NLCP and two of decrease, all are statistically significant at either 95% or at 90% confidence level (See Table 1). These findings show that dependence on locally created knowledge is increasing for all assignee types except the federal government. The outlying case of the federal government does not detract from our findings. The federal government s research activities are less subject to market forces than other economic actors, and less so even than universities (who must compete for their research funding). Federal government research also might depend less on locally-created knowledge because it allocates its resources through national political processes and bureaucracies, and to some extent it is insulated from time and price pressures present for other assignee types. However, more research on the geography of government research is needed. Table 2. t-test Results of Changes in Ratio of Net Local Citations by Assignee Type Our findings lend partial support to Hypothesis 2. As seen in Figure 3, the NLCP of different assignee types are considerably different and the difference between assignee types is persistent over three geographical levels. At all three geographical

23 Increasing Importance of Proximity in Innovation 23 levels, inventors in the categories of unassigned and individual have greater than average dependency on locally created knowledge. On the other hand, top 500 firms and other firms are below average in this regard, as expected. While large firms may be able to overcome distance in communication by investing in building contact networks, an independent inventor or small entity is likely to be more confined to nearby sources. The difference between Top 500 firms and other firms is unclear, however. Figure 3 shows that the difference between these two categories is not systematic at any of three geographical levels. In some years, Top 500 firms are more dependent on locally created knowledge while for other years, they are less dependent. The abovementioned assignee categories show consistent results across geographic scales, but this is not the case for universities. At the country level, universities depend more on domestically created knowledge than the other assignee categories. However, this is not the case at state and metropolitan levels. In other words, the nationality of a university is important in acquiring knowledge, but the location of university inside a country does not matter so much. We can speculate as to why this might be the case. University research tends to be more basic than research by firms or by independent inventors. Moreover, there are rarely sufficient colleagues in a given specialized area in a single research unit, putting a premium on communication with colleagues in other parts of the country via academic journals, national conferences and other methods. International communication is more limited, due to cost, language and familiarity constraints. Given, however, that basic science (especially in the natural sciences and engineering) uses a universal technological language, we found this result to be surprising and in need of further explanation.

24 Increasing Importance of Proximity in Innovation 24 Some Implications for Research and Policy One of the most important and earliest empirical claims that multinational enterprises appeared to be deeply linked, for their most innovative (patented) activities, to their home territories was made by Pavitt and Patel in their widely-cited 1991 article. Since then, there has been considerable empirical confirmation of the effect of geographical proximity in innovation, through the research of Jaffe, Trajtenberg and Henderson (1993), Feldman and Florida (1994) and Anselin, Varga and Acs (1997), Audretsch (1998) and most recently Baptista (2000). Until now, however, there has been no indication of the time trend of this phenomenon, i.e. whether it has increased with the appearance of new forms of organization of the innovation process and with the complex effects of ICTs on the contemporary economy. Hence, it has been difficult to weigh the importance of findings that geographical proximity exists. Our research adds this critical missing element, finding that inventors cite local patents increasingly over time. This is confirmed for all assignee categories except the federal government: Top 500 Most Inventive Businesses; Other Businesses; Universities; Individuals; Unassigned. In other words, in spite of the increasing global flow of ideas, proximity appears to have become even more important than it was in 1975, and a threshold was crossed in the late 1980s or early 1990s toward greater geographically-localized citation activity. There are some differences between assignee categories. Unassigned and Individuals were more dependent on locally created knowledge than businesses but the Pavitt-Patel effect is still present for businesses and it is increasing over time along

25 Increasing Importance of Proximity in Innovation 25 with the other assignee categories. Our finding thus calls for nuancing the claim that these very large businesses are somehow able to internalize complex knowledge flows and extend them over large distances, and even to do away with the effect of locality in their innovative processes. A number of authors have advanced the notion that localized interaction in innovation could generate geographically-differentiated technological trajectories, implying that there are strong geographical barriers to, and lags in, the technological diffusion process, and evolutionary feedbacks within geographically-bounded environments. This idea, of course, goes against the grain of long-term technological convergence through imitation and adoption, or at least suggests that the world, at any given moment, is in an out-of-equilibrium state when it comes to technology (Rigby and Essletzbichler 1997). Our analysis cannot confirm or reject such wider notions, but it does suggest that more research is merited on the relationship between the geographically-uneven and partially-bounded creation of technology and the geography of technology diffusion are key defining aspects of technological evolution as a whole, dynamics that are much more complex than most models tend to suggest. Nothing in this analysis is intended to deny that communication and transportation technologies are making long distance interactions easier than ever. Rather, we believe the the demise of distance argument and our emphasis on the importance of proximity are complementary in understanding the contemporary geography of innovation. This is because each argument deals with a different dimension of the innovation process. On one hand, there is no question that today s scientists and engineers can

26 Increasing Importance of Proximity in Innovation 26 communicate at long-distance more easily (Johnson, Siripong and Brown, 2002). Located in different parts of the world, Route 128 in Massachusetts, Silicon Valley in California, Shinju Science Park in Taiwan, Bangalore Software Park in India and other high-tech centers of the world are all well-connected with one another. Compared to fifty years ago, when the high tech centers were close to each other, principally in the Northeast and Midwest of the U.S, the distance between high-tech centers are surprisingly long. In this sense, there is an ongoing death of distance in that the centers of knowledge creation are farther and farther away from one another. On the other hand, the findings here lend additional support to the notion that there are constraints to long-distance interaction and consequent advantages to proximity, in the creation of certain types of patentable knowledge, and that these may be increasing over time. There is now a considerable literature on this issue, focusing on various hypotheses about the precise nature of such constraints, including our own earlier contributions to that debate (Storper and Venables, 2003; Storper, 1997). In any case, it may be that certain innovation processes involve both globalization and localization. High-tech centers located at great distance are well connected but they might be specialized in different technologies, with knowledge spillovers inside the centers becoming increasingly important. However, in order to answer this question with any assurance, further research will need to disaggregate the dynamics described here, and especially to do so on a filière or technology field level. This paper s findings also imply that if proximity plays an increasingly important role in the innovation process, then there is room for additional reflection about whether public policy should take this explicitly into account. Most technology policies

27 Increasing Importance of Proximity in Innovation 27 are meant, by design, to be indifferent to location, precisely because they want to stimulate knowledge diffusion, competition, and avoid generating rents, although certain aspects of US federal policy toward knowledge creation (especially in agriculture) have a history of funding localized knowledge creation and then encouraging its non-localized diffusion (Storper, 1995a,b). Along these lines, there appears to be some reason to think that local governments who attempt to attract inward investment should privilege those activities that can create synergy with existing industries, regardless of whether the resulting rents would be captured locally or not. In any event, the evidence presented here, that the importance of localized technological interactions are increasing and not decreasing, and that this is true of even the very biggest firms, should now be taken into account in assessing the virtues of different types of policy approach. References Acs, Zoltan J., 2002, Innovation and the Growth of Cities, Cheltenham: Edward Elgar. Acs, Zoltan J., David B. Audretsch, and Maryann P. Feldman Real effects of academic research: Comment. American Economic Review March Anselin, L., A. Varga, and Z. Acs Local geographic spillovers between university research and high technology innovations. Journal of Urban Economics 42 (3): Audretsch, D. B Agglomeration and the location of innovative activity. Oxford Review of Economic Policy 14 (2): Audretsch, David B., and Maryann P. Feldman R&D spillovers and the geography of innovation and production. American Economic Review June 630. Baptista, R Do innovations diffuse faster within geographical clusters? International Journal of Industrial Organization 18 (3): Baptista, R., and P. Swann Do firms in clusters innovate more? Research Policy 27 (5): Braczyk, Hans-Joachim, Philip Cooke, and Martin Heidenreich Regional innovation systems : the role of governances in a globalized world. London: UCL Press.

28 Increasing Importance of Proximity in Innovation 28 Breschi, Stefano, and Francesco Lissoni Knowledge spillovers and local innovation systems: a critical survey.castellanza, Italy?Cattaneo University. Breschi, Stefano, and Franco Malerba Networks of knowledge through patent citations: An analysis of EPO patent citations data Paper presented at the DRUID Summer Conference June 2003 Available from Breschi, Stefano, Francesco Lissoni, and Franco Malebra Knowledge networks from patent citations?-methodological issues and preliminary results Paper presented at the DRUID Summer Conference June 2003 Available from Cairncross, Frances The death of distance: How the communications revolution will change our lives.london?orion. Cantwell, J., and S. Iammarino EU regions and multinational corporations: Change, stabiloity,and strengthening of technological comparative advantage. Industrial and corporate change 10: Cooke, Philip and Kevin Morgan The network paradigm - New departures in corporate & regional development. Environment & Planning D: Society and Space 8(1): 7-34 Cooke, Philip, and Kevin Morgan eds The associational economy : firms, regions, and innovation. Oxford England ; New York: Oxford University Press. Dicken, Peter 1992 Global Shift: The Internationalization of Economic Activity 2nd edition, New York : Guilford Press Dunning, John H.ed Regions, globalization, and the knowledge-based economy. Oxford: Oxford Press. Feldman, Maryann P The geography of innovation. Boston: Kluwer Academic Publishers. Feldman, Maryann P., and Richard Florida The geographic sources of innovation: technological infrastructure and product innovation in the United States. Annals of the Association of American Geographers June 210 Fischer, Manfred M., Javier Revilla Diez, Folke Snickars, J. Revilla Diez Metropolitan Innovation Systems: Theory and Evidence from Three Metropolitan Regions in Europe New York : Springer Foray, D The secrets of industry are in the air: Industrial cooperation and the organizational dynamics of the innovative firm' Research Policy 20 : , Glaeser, E. L Learning in cities. Journal of Urban Economics 46 (2): Hakanson, Lars Epistemic communities and cluster dynamics: On the role of knowledge in industrial districts Paper presented at the DRUID Summer Conference June 2003 Available from